Hi Miao,
As you suggested, in btrfs_recover_log_trees(), the items to modify in the
transaction are
not known before entering a tree, we can use the global block reservation for
it.
Signed-off-by: Itaru Kitayama <kitay...@cl.bb4u.ne.jp>
---
fs/btrfs/tree-log.c | 2 ++
1 files changed, 2 insertions(+), 0 deletions(-)
diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c
index 054744a..7df8c7b 100644
--- a/fs/btrfs/tree-log.c
+++ b/fs/btrfs/tree-log.c
@@ -3081,6 +3081,8 @@ int btrfs_recover_log_trees(struct btrfs_root
*log_root_tree)
trans = btrfs_start_transaction(fs_info->tree_root, 0);
+ trans->block_rsv = &fs_info->global_block_rsv;
+
wc.trans = trans;
wc.pin = 1;
--
1.7.3.4
On Thu, 06 Jan 2011 14:47:41 +0800
Miao Xie <mi...@cn.fujitsu.com> wrote:
> Hi, Kitayama-san
>
> Firstly, thanks for your test.
>
> On Sat, 1 Jan 2011 00:43:41 +0900, Itaru Kitayama wrote:
> > Hi Miao,
> >
> > The HEAD of the perf-improve fails to boot on my virtual machine.
> >
> > The system calls btrfs_delete_delayed_dir_index() with trans block_rsv set
> > to NULL,
> > thus selects, in get_block_rsv(), empty_block_rsv whose reserve is 0 (and
> > size is also 0),
> > which leads to ENOSPC. I wonder below patch is enough reserve metadata to
> > finish
> > btrfs_recover_log_trees() without going to ENOSPC. I appreciate your review.
> >
> > Singed-Off-by: Itaru Kitayama<kitay...@cl.bb4u.ne.jp>
> >
> > diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c
> > index 054744a..f26326b 100644
> > --- a/fs/btrfs/tree-log.c
> > +++ b/fs/btrfs/tree-log.c
> > @@ -3079,7 +3079,7 @@ int btrfs_recover_log_trees(struct btrfs_root
> > *log_root_tree)
> > path = btrfs_alloc_path();
> > BUG_ON(!path);
> >
> > - trans = btrfs_start_transaction(fs_info->tree_root, 0);
> > + trans = btrfs_start_transaction(fs_info->tree_root, 4);
>
> I don't think this change is right, because we don't know how many leaves we
> may change
> when doing log tree replay, so we can't set the secondargument to 4.
>
> And I think the original code is right, because the space reservation is used
> to avoid filesystem
> operations being broken by that other users hogging all of the free space.
> but this function is
> invoked when we mount a filesystem, at this time, no other user can access
> the filesystem,
> so we can use all of the free space, thus we needn't reserve any free space
> for log tree replay.
>
> I don't understand the log tree very well, maybe there is something wrong
> with what I said.
> If what I said above is right, we should look for another way to fix this
> problem.
>
> (I'm making the second version of this patchset now, I'll fix it in it. So if
> your patch is right,
> I'll want to add it into my patchset.)
>
> Thanks again for your test.
> Miao
>
> >
> > wc.trans = trans;
> > wc.pin = 1;
> >
> > Here's the log:
> >
> > kernel BUG at fs/btrfs/tree-log.c:678!
> > invalid opcode: 0000 [#1] SMP
> > last sysfs file: /sys/devices/virtual/bdi/btrfs-1/uevent
> > CPU 1
> > Modules linked in: floppy mptspi mptscsih mptbase scsi_transport_spi [last
> > unloaded: scsi_wait_scan]
> >
> > Pid: 308, comm: mount Not tainted 2.6.36-perf-improve+ #1 440BX Desktop
> > Reference Platform/VMware Virtual Platform
> > RIP: 0010:[<ffffffff811eb161>] [<ffffffff811eb161>]
> > drop_one_dir_item+0xd6/0xfb
> > RSP: 0018:ffff88007a5a5858 EFLAGS: 00010286
> > RAX: 00000000ffffffe4 RBX: ffff88007d2b7800 RCX: ffff880037e8b240
> > RDX: 0000000000000000 RSI: ffffea0000c3ae68 RDI: 0000000000000206
> > RBP: ffff88007a5a58c8 R08: 00000000005e6760 R09: ffff88007a5a55e8
> > R10: ffff88007a5a55e0 R11: ffff88007a5a5648 R12: ffff880037e8b120
> > R13: ffff880037e98cc0 R14: ffff88007b371c90 R15: 0000000000000005
> > FS: 00007f37b63c4800(0000) GS:ffff880002040000(0000) knlGS:0000000000000000
> > CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > CR2: 00007f37b55f0190 CR3: 000000007a4d9000 CR4: 00000000000006e0
> > DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> > DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
> > Process mount (pid: 308, threadinfo ffff88007a5a4000, task ffff88007a5c9720)
> > Stack:
> > 0000000000000005 0000000000000d75 ffff880037e98550 ffff880037dcf000
> > <0> 000000000016e730 0000000000000001 0000000000000100 00000000005e7b60
> > <0> ffff88007a46d000 ffff880037e8b120 ffff88007d2b7800 ffff880037e98550
> > Call Trace:
> > [<ffffffff811ec339>] add_inode_ref+0x32a/0x403
> > [<ffffffff811ec59a>] replay_one_buffer+0x188/0x209
> > [<ffffffff811bafef>] ? verify_parent_transid+0x36/0xf9
> > [<ffffffff811e8eb9>] walk_up_log_tree+0x109/0x1d1
> > [<ffffffff811ec412>] ? replay_one_buffer+0x0/0x209
> > [<ffffffff811e930f>] walk_log_tree+0x9b/0x187
> > [<ffffffff811eaf73>] btrfs_recover_log_trees+0x18a/0x2a2
> > [<ffffffff811ec412>] ? replay_one_buffer+0x0/0x209
> > [<ffffffff811bb123>] ? btree_read_extent_buffer_pages+0x71/0xaf
> > [<ffffffff811becfe>] open_ctree+0xf8f/0x12c6
> > [<ffffffff811a69b4>] btrfs_get_sb+0x225/0x459
> > [<ffffffff810fe143>] ? __kmalloc_track_caller+0x13a/0x14c
> > [<ffffffff8110d458>] vfs_kern_mount+0xbd/0x1a7
> > [<ffffffff8110d5aa>] do_kern_mount+0x4d/0xed
> > [<ffffffff8112318e>] do_mount+0x754/0x7cb
> > [<ffffffff810dae6d>] ? memdup_user+0x60/0x80
> > [<ffffffff810daecb>] ? strndup_user+0x3e/0x54
> > [<ffffffff8112328d>] sys_mount+0x88/0xc2
> > [<ffffffff81009c72>] system_call_fastpath+0x16/0x1b
> > Code: de e8 8f e5 ff ff 85 c0 74 04 0f 0b eb fe 48 8b 55 a0 48 8b 7d a8 45
> > 89 f9 4d 89 f0 4c 89 e9 48 89 de e8 65 bf fd ff 85 c0 74 04<0f> 0b eb fe 4
> > RIP [<ffffffff811eb161>] drop_one_dir_item+0xd6/0xfb
> > RSP<ffff88007a5a5858>
> > ---[ end trace 2ec638d9e30d6102 ]---
> >
> >
> > On Wed, 01 Dec 2010 16:09:17 +0800
> > Miao Xie<mi...@cn.fujitsu.com> wrote:
> >
> >> Compare with Ext3/4, the performance of file creation and deletion on
> >> btrfs is
> >> very poor. the reason is that btrfs must do a lot of b+ tree insertions,
> >> such as
> >> inode item, directory name item, directory name index and so on.
> >>
> >> If we can do some delayed b+ tree insertion or deletion, we can improve the
> >> performance, so we made this patch which implemented delayed directory name
> >> index insertion and deletion.
> >>
> >> Beside that, we found we must map the page every time we want to set a
> >> member
> >> variable of the inode item, it is inefficient. We just do it at first to
> >> reduce
> >> the times of mmap(). By this way, we can also improve the performance of
> >> file
> >> creation and deletion.
> >>
> >> I did a quick test by the benchmark tool[1] and found we can improve the
> >> performance of file creation by ~11%, and file deletion by ~14%.
> >>
> >> Before applying this patch:
> >> Create files:
> >> Total files: 50000
> >> Total time: 1.188547
> >> Average time: 0.000024
> >> Delete files:
> >> Total files: 50000
> >> Total time: 1.662012
> >> Average time: 0.000033
> >>
> >> After applying this patch:
> >> Create files:
> >> Total files: 50000
> >> Total time: 1.057432
> >> Average time: 0.000021
> >> Delete files:
> >> Total files: 50000
> >> Total time: 1.422851
> >> Average time: 0.000028
> >>
> >> You can also try out the patchset by pulling:
> >> git://repo.or.cz/linux-btrfs-devel.git perf-improve
> >>
> >> [1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3
> >>
> >> ---
> >> fs/btrfs/Makefile | 2 +-
> >> fs/btrfs/btrfs_inode.h | 2 +
> >> fs/btrfs/ctree.c | 13 +-
> >> fs/btrfs/ctree.h | 21 +-
> >> fs/btrfs/delayed-dir-index.c | 790
> >> ++++++++++++++++++++++++++++++++++++++++++
> >> fs/btrfs/delayed-dir-index.h | 92 +++++
> >> fs/btrfs/dir-item.c | 61 +++-
> >> fs/btrfs/extent-tree.c | 21 ++
> >> fs/btrfs/inode.c | 189 +++++++---
> >> fs/btrfs/transaction.c | 9 +
> >> fs/btrfs/transaction.h | 2 +
> >> 11 files changed, 1117 insertions(+), 85 deletions(-)
> >> --
> >> To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
> >> the body of a message to majord...@vger.kernel.org
> >> More majordomo info at http://vger.kernel.org/majordomo-info.html
> >>
> >
> >
>
>
--
Itaru Kitayama <kitay...@cl.bb4.ne.jp>
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
